Members of the LIDO Study Group are listed in the acknowledgments.
ClinicalTrial.gov identifier number: NCT00492700.
Conflict of interest: This was an investigator initiated trial and GlaxoSmithKline Pharmaceuticals had no direct or indirect involvement in the design of the trial, data collection or preparation, or submission of the article. GSK provided rosiglitazone for this trial and partly funded the trial. None of the authors has a personal conflict of interest with the manufacturers of any of the marketed glitazones. V.R. is a consultant to Astellas, Axcan, Gilead, Genentech, Intercept, and Sanofi-Aventis. T.P. is a consultant for and owns 15% of Biopredictive, a company that markets FibroTest and SteatoTest. These conflicts of interest have been disclosed to study participants.
Short-term trials of glitazones in nonalcoholic steatohepatitis (NASH) yielded controversial histological results. Longer treatment might result in additional improvement. After a 1-year randomized trial, 53 patients underwent a control liver biopsy and were enrolled in an open-label extension trial of rosiglitazone (RSG), 8 mg/day for 2 additional years. In all, 44 completed the extension phase including 40 with a third liver biopsy. Of these, 22 received placebo (PLB) in the randomized phase (PLB-RSG), and 18 RSG (RSG-RSG). During the 2-year extension phase serum insulin decreased by 26%, homeostasis model assessment (HOMA) by 30%, and alanine aminotransferase (ALT) by 24%. However, there was no significant change in the mean NASH activity score (NAS) (3.8 ± 2.11 versus 3.68 ± 1.8), ballooning score, fibrosis stage (1.76 ± 1.18 versus 1.85 ± 1.19), or area of fibrosis by micromorphometry (4.43% ± 0.68 to 5.54% ± 0.68). In the PLB-RSG group steatosis significantly decreased after 2 years of RSG (median decrease of 15%); in the RSG-RSG group, after an initial decline in the first year of 20%, 2 additional years of RSG did not result in further improvement. Likewise, there was no improvement in the NAS score, ballooning, intralobular inflammation, fibrosis stage, or area of fibrosis with 2 additional years of RSG in the RSG-RSG group. Conclusion: Rosiglitazone has a substantial antisteatogenic effect in the first year of treatment without additional benefit with longer therapy despite a maintained effect on insulin sensitivity and transaminase levels. This suggests that improving insulin sensitivity might not be sufficient in NASH and that additional targets of therapy for liver injury should be explored. (HEPATOLOGY 2009.)
Nonalcoholic steatohepatitis (NASH) is an increasingly frequent liver disease in patients with metabolic risk factors and insulin resistance such as overweight/obesity, visceral adiposity, diabetes, and dyslipidemia.1, 2 Although the majority of afflicted patients have a mild hepatic course, a significant proportion do progress to cirrhosis, hepatocellular carcinoma, and endstage liver disease, with a resultant increase in liver-related mortality and overall mortality.3–6 Therefore, the discovery of effective treatments for this disease is a major challenge in clinical hepatology.
Currently there is no effective treatment in the majority of NASH patients who fail to respond to or implement diet and lifestyle changes. Thiazolidinediones (glitazones) are attractive drug candidates, because their multiple insulin-sensitizing actions counteract insulin resistance, the almost universal underlying metabolic disturbance in NASH. A few randomized, controlled trials with either rosiglitazone or pioglitazone have been reported, with a variable effect on liver histology.7–10 Although all studies have convincingly demonstrated an improvement in steatosis, not all of them have reported a concomitant improvement in inflammation and liver cell injury (defining steatohepatitis) and only one has reported a marginal improvement in fibrosis.10 Glitazones, however, do not directly target the liver, at least in humans. Instead they promote preadipocyte differentiation into small, insulin-sensitive adipocytes; they induce a redistribution of fat from ectopic sites such as the liver and muscle to the peripheral subcutaneous adipose tissue; and they correct adipocytokine imbalance, promoting an insulin-sensitive profile, mainly by increasing circulating levels of adiponectin.11 Because improvement in liver injury might only be subsequent to the modification of the metabolic milieu, it can arguably occur with significant delay and require longer treatments. In that regard, all studies published to date were only 6 or 12 months long, a rather short treatment period that might be insufficient to allow for a full histological improvement.
We designed the current Fatty Liver Improvement by Rosiglitazone Therapy (FLIRT 2) study as an open-label extension study of the FLIRT, randomized, placebo-controlled trial of rosiglitazone in NASH patients.9 The aim of this extension study was to determine whether prolonged therapy with rosiglitazone is associated with further histological improvement, and, in particular, if fibrosis regression can be obtained.
AF, area of fibrosis; ALT, alanine aminotransferase; FLIRT: Fatty Liver Improvement by Rosiglitazone Therapy; HOMA, homeostasis model assessment; NAS, NASH activity score; NASH, nonalcoholic steatohepatitis; PLB, placebo; RSG, rosiglitazone.
Patients and Methods
In the FLIRT Trial, 63 patients with histologically proven NASH and increased transaminase values (detailed inclusion criteria in Ref.9) were randomized to receive rosiglitazone (8 mg /day, n = 32) or placebo (n = 31) for 1 year (M0 to M12); they then underwent an additional evaluation at M16 after a 4-month follow-up off therapy. With the double-blinding still in effect, patients were enrolled at M16 in an open-label, extension trial, where they received rosiglitazone, 8 mg per day (4 mg the first month) for 2 additional years up to M40 (FLIRT 2 trial). The flow chart of study participation is shown in Fig. 1. Seven patients refused to be included in the open-label extension phase, and three patients were not solicited to participate (investigator's concern about long-term compliance with study medication and constraints of the protocol). Thus, 53 patients entered the extension phase; 25 of these received rosiglitazone in the randomized phase and 28 received placebo. Patients were followed every 4 months with clinical visits and biological data. A third liver biopsy was performed at M40 at the end of the 2-year extension phase. All patients gave written informed consent and the study protocol was approved by the Ethics Committee of Pitié-Salpêtrière Hospital.
Liver samples were paraffin-embedded and stained with hematoxylin-eosin-safran and Sirius red. Steatosis was semiquantitatively assessed as the percentage of hepatocytes containing fat droplets within a lobule (0%-100%, steatosis score). Fibrosis staging was performed according to Brunt et al.12: F0 = absence of any fibrosis; F1 = isolated perisinusoidal or portal/periportal fibrosis; F2 = perisinusoidal and portal/periportal fibrosis; F3 = bridging fibrosis; F4 = cirrhosis. The NASH Activity Score (NAS) was calculated according to Kleiner et al.13 The effect of treatment was assessed on both overall fibrosis and, specifically, perisinusoidal fibrosis. All pathological analyses were performed by a single pathologist, unaware of the clinical data or the sequence of treatment assignment. The reading of the three biopsies for all patients was performed in random order; elementary histological lesions were scored at the baseline (M0), 1-year (M12), and 3-year (M36) liver biopsies.
Hepatic fibrosis was also quantified by micromorphometry. The image of a whole tissue section was captured at ×10 magnification with a high-resolution numerical camera (Leica DFC320) coupled with a light microscope. The area of fibrosis (AF) was measured after Sirius red staining of 4-μm tissue sections using an image analyzer computer system (Leica Q500 iW, Cambridge, UK). A special histomorphometry program was created with interactive software QUIPS (Quantimet Image Processing Software, Leica) for this study. The AF was expressed as percentage of the total area of the tissue section.
Clinical and Biochemical Data.
Diabetes was considered present if patients were taking medications for diabetes (metformin or sulfamides) or if fasting serum glucose was ≥7 mmol/L in untreated patients. Arterial hypertension was considered present if patients were taking antihypertensive drugs. All laboratory tests at all timepoints were obtained from a centralized laboratory after a 12-hour overnight fast. The homeostasis model assessment (HOMA) index, a marker of insulin resistance, was calculated as: glucose (mmol/L) × insulin (μIU/L)/22.5. HOMA estimates of insulin resistance based on the measurement of serum insulin and glucose values can be used in patients with type 2 diabetes who are not treated with insulin.14 Numerous therapeutic trials including pivotal trials of rosiglitazone either in monotherapy15–18 or in combination with metformin19 reported on improvement in serum insulin15, 16, 18 or HOMA estimates of insulin sensitivity15, 17, 19 in patients with type 2 diabetes not receiving insulin therapy.
Numerical variables were compared using analysis of variance for those that were normally distributed and nonparametric tests such as the Mann-Whitney U test for those without a normal distribution. Proportions were compared using Fisher's exact test. Correlations between efficacy endpoints and numerical clinical and biologic variables were tested by the Spearman rank-correlation matrix.
In this study we specifically report on changes in laboratory parameters during the extension phase of the trial and expressed the values as percentage change between the initiation of the extension phase (M16) and the end of study (M40). No formal power calculations were made, as most patients were invited to proceed in the long-term extension trial.
Fifty-three patients were included in the extension phase, which started at M16. The clinical characteristics of these patients were compared to those of the 10 patients who were not included (Table 1). The only difference between these two groups was a lower serum level of high-density lipoprotein (HDL) cholesterol in the included patients. The two groups were also comparable in terms of severity of histological injury on the first biopsy (M0) (Table 1), as well as on the second biopsy at M12, i.e., 4 months before inclusion in the extension phase (data not shown).
Table 1. Characteristics of Patients Included or Not in the Extension Trial
Likewise, there were no differences at inclusion in the extension trial (M16) between patients who received placebo (PLB) and those who received rosiglitazone (RSG) during the randomized trial (Table 1).
Among the 53 patients included, five patients discontinued the study for side effects and four patients were lost to follow-up (two at M24, one at M26 and M38). Thus, 44 patients completed the additional 2-year treatment period (23 in the PLB-RSG group and 21 in the RSG-RSG group).
Rosiglitazone induced a significant decrease in the mean alanine aminotransferase (ALT) values between M16 and M40 in the 44 patients who completed the extension trial: from 66 (standard deviation [SD] 45) IU/L at M16 to 45 (SD 33) IU/L at M40 (P = 0.01). This corresponds to an average 25% reduction, which is comparable to the 30% reduction observed in the RSG arm during the randomized phase of the trial9 (Fig. 2). Therefore, the efficacy of RSG for the reduction in ALT levels was not modified by previous treatment courses or the rebound following the 4-month follow-up period. On the other hand, the maximum ALT reduction was obtained in the first 4-6 months of therapy. ALT levels were normal at M40 in 20 patients (45%). Figure 2 shows the mean serum ALT course during the randomized and the extension phases of the trial in the PLB-RSG and the RSG-RSG groups of patients. Of note, there was no breakthrough during therapy, indicating maintained efficacy throughout the whole treatment period.
RSG induced a reduction of the median HOMA score from 3.48 (interquartile range [IQR] 3.37) at M16 to 2.51 (2.53) at M40 (P = 0.017) (see Supporting Fig. 1 on the website of the journal for the kinetics of changes in HOMA score) and of the median fasting insulin levels from 13.9 (10.9) μIU/L at M16 to 10.2 (8.1) μIU/L at M40 (P < 0.008) without a change in serum glucose or hemoglobin (Hb) A1c levels. Table 2 shows that the magnitude of reduction during therapy was 30% in HOMA levels and 26% in serum insulin levels. This magnitude of improvement in serum insulin and HOMA scores is comparable to reported trials of RSG in monotherapy in patients with type 2 diabetes not treated with insulin17 (see Patients and Methods section). There was a 13% but nonsignificant reduction in serum triglyceride levels, whereas HDL cholesterol was unchanged during therapy.
Table 2. Changes in Hepatic and Metabolic Parameters During the 2-Year Extension Phase
During the extension trial the histological score of steatosis was not significantly reduced in the overall population (40 patients enrolled in the extension trial) between M12 (score of 38%, SD 21) and M40 (score of 32%, SD 16), P = 0.16, which contrasts with the strong antisteatogenic effect of RSG in the randomized phase of the trial. In the same population, when the histological score of steatosis at inclusion (M0) was compared with that at M40, there was a significant reduction: from 50% (SD 20) at M0 to 32% (SD 16) at M40, P < 0.001. This strongly suggests that most of the antisteatogenic effect was obtained during the first year of therapy. This was further demonstrated by comparing the course of liver fat on serial biopsies in the two groups of patients according to the treatment received during the randomized phase of the trial. Figure 3 shows that in patients initially treated with RSG (RSG-RSG group), the significant reduction in steatosis obtained after 1 year of treatment (median −20%, IQR 30) was not followed by any additional decrease thereafter. In contrast, the PLB-RSG group displayed a nonsignificant variation of steatosis while on placebo followed by a significant reduction when treated with RSG. Moreover, the magnitude of decrease in the steatosis score in the PLB-RSG group while on rosiglitazone for 2 years (median −15% [IQR 30%]) was similar to that in the RSG-RSG group obtained during the first year of RSG (median −20% [IQR 30%]).
There was a nonsignificant reduction in the mean NAS score during the 2-year extension phase in the overall population: 3.8 (SD 2.11) at M12 versus 3.68 (SD 1.8) at M40. The ballooning score increased nonsignificantly from 0.98 (SD 0.8) to 1.13 (SD 0.65), P = 0.36, whereas the intralobular inflammation/necrosis score did not change (1.05, SD 0.75 at M12 versus 1.05, SD 0.78 at M40). Table 3 details the scores for the different components of these lesions on the three successive liver biopsies according to the treatment regimen. In contrast to its antisteatogenic effect, RSG did not induce a significant reduction in the score of liver cell damage/inflammation after 1, 2, or 3 years of treatment. Improvement by one point or more was seen in seven patients for ballooning, in eight patients for lobular necroinflammation, and in 13 patients for the NAS score. Worsening by one point or more occurred in 12, 10, and 12 patients, respectively.
Table 3. Histological Changes During Therapy in the 40 Patients of the Extension Trial According to the Initial Treatment Allocation (Mean, SD)
PLB-RSG (n = 22)
RSG-RSG (n = 18)
In the overall population, there was a nonsignificant increase in the mean global fibrosis score between M12 and M40 while on RSG (1.76, SD 1.18 at M12 versus 1.85, SD 1.19 at M40, P = 0.74). There was no significant change in the perisinusoidal fibrosis score: 1.23, SD 0.66 at M12 versus 1.18, SD 0.68 at M40, P = 0.74. Table 3 shows no clear pattern of change in the different fibrosis scores on the three sequential liver biopsies according to the treatment regimen. The fibrosis score improved by one point or more in nine patients (and worsened in 11 by the same amount) and the perisinusoidal fibrosis score improved in 11 patients but worsened in 9 of the 40 patients.
In the first year, during the randomized phase of the trial, the mean area of fibrosis decreased from 4.22 (SD 3.3) to 3.74 (SD 2.68) in the 32 patients who received RSG and increased from 4.22 (SD 2.44) to 5.55 (3.87) in the 31 patients who received PLB. Neither variation was statistically significant (P = 0.52 and 0.12, respectively). Supporting Fig. 1 shows that extending RSG treatment for 2 additional years did not result in an antifibrotic effect in the RSG-RSG group nor did it reverse fibrosis progression in the PLB-RSG group.
Histological Changes in Relation to Metabolic Improvement
Because glitazones are able to improve both insulin resistance and steatosis, we then tested the hypothesis that with long-term treatment a marked improvement in one or both of these conditions will result in an improvement in necroinflammatory and/or fibrotic features. We defined improvement in insulin sensitivity as a 40% or more reduction in the HOMA score between M12 and M40. Table 4 shows the mean reduction in the scores of inflammatory and fibrotic lesions between the 14 patients with a significant improvement in insulin sensitivity and the 26 patients with less or no improvement. Improvement in insulin sensitivity did not result in a reduction in fibrotic lesions, nor in a clear trend in necroinflammatory lesions except for a marginal effect on hepatocyte ballooning. Likewise, a major improvement in steatosis, defined as a reduction of 30% or more in the steatosis score9 between M12 and M40, did not result in a significant reduction in fibrosis. There was, however, a significant improvement in lobular necroinflammation associated with the improvement in steatosis, as well as a trend of an improvement in hepatocyte ballooning. Of note, the median ALT reduction was higher in patients with a major improvement in insulin sensitivity (0.61 [IQR 0.39] versus 0.78 [0.41], P = 0.03) and in those with a major reduction in steatosis (0.49 [0.62] versus 0.76 [0.37], P = 0.06, respectively).
Table 4. Changes in Histological Lesions According to Improvement in Steatosis or Insulin Resistance
Improvement in Steatosis Score
Improvement in HOMA Score
<30% (N = 31)
>30% (N = 9)
<40% (N = 26)
>40% (N = 14)
Numbers represent mean changes (SD) for each elementary histological lesion or score during the open-label extension phase (differences between the M40 and the M12 biopsy). Negative values denote improvement.
The main side effects are shown in Table 5. As anticipated there was a median weight gain of 2 kg, with a steady increase throughout the entire 2-year period of the study. However, weight gain was not universal, as 11 patients lost 2 kg or more. Asthenia, muscular cramps, and swollen legs were also frequently reported, albeit intermittently. Among the 14 patients (seven men, seven women) with muscle cramps, only five were treated with statins or fibrates. Mild exertional dyspnea, without any objective clinical or radiological sign of cardiac insufficiency, was also reported in two patients. There were no cardiovascular events during the trial, although two patients experienced a nonfatal myocardial infarction after the study was completed. There were no episodes of hepatotoxicity.
Table 5. Adverse Events Reported at Any Time and at Least Once During the Study
Occurrence/Frequency n (%)
Weight gain, median, kg [IQR]
Weight gain >3 kg
Five patients withdrew from the study, one for planned antiobesity surgery, one for personal reasons, and three specifically for side effects: swollen legs and gastrointestinal symptoms: nausea and bloating. The latter symptoms disappeared quickly after discontinuation of the drug but were considered unrelated to the study drug by the investigator.
In a previous, 1-year, randomized, placebo-controlled trial of RSG in patients with NASH, we demonstrated a marked antisteatogenic effect of RSG together with a significant reduction in transaminase values. However, there was no clear improvement in necroinflammatory lesions and, as in other reports,7, 8, 20 there was no effect on liver fibrosis. Glitazones first need to improve insulin resistance before any potentially beneficial hepatic effect might ensue. Therefore, it can be argued that longer treatment exposure is necessary if inflammation, liver cell injury, and the fibrotic response are to be improved as a consequence of the correction of the underlying metabolic condition.
This open-label extension study tested whether prolonged exposure to RSG for a total of 2 or 3 years provides greater improvement of liver injury than a 1-year course. The design of the trial included an intermediary biopsy that allowed assessment of whether the histological improvement is a gradual and continuing process that is proportional to the length of exposure to the drug or if it is obtained early on with no further improvement beyond 1 year of treatment. This study clearly shows that RSG has a robust and reproducible antisteatogenic effect, with a reduction in transaminase values that is correlated with the loss of liver fat. However, reduction in liver fat occurs mostly in the first year of treatment, with no further reduction thereafter. Indeed, patients who received RSG for 3 years had a strong reduction in steatosis on the year 1 biopsy with no further reduction later, at year 3 biopsy. Also, patients initially assigned to the PLB group who were treated with RSG for the subsequent 2 years lost as much liver fat in those 2 years as did those initially assigned to RSG in the first year. The reasons for the exhaustion of the antisteatogenic effect are not clear at the present time. Interestingly, the improvement in insulin sensitivity mirrored the antisteatogenic effect: it reached a maximum level in the first year with no further enhancement subsequently. Therefore, this might simply reflect the fact that glitazones have a significant but still limited insulin-sensitizing potency and raises the question as to whether drugs with a stronger insulin-sensitizing effect might result in a more drastic reduction of steatosis.
The biochemical transaminase response was not affected by prior treatment with RSG and by the rapid relapse during the short off-treatment phase between M12 and M16. Moreover, the effect on aminotransferases was maintained without breakthrough throughout the whole treatment period. Interestingly, the magnitude of the reduction in transaminase levels was comparable after a short 1-year period of treatment and a longer additional 2-year period. This follows the same pattern described above for the reduction in liver fat and the improvement in insulin resistance. Moreover, there was a significant correlation between these three parameters. It is therefore tempting to speculate that improvement in insulin sensitivity, loss of liver fat, and reduction in serum aminotransferases are all part of an interconnected pattern of response to glitazone therapy.
In contrast to steatosis and transaminase levels, there was no consistent and beneficial effect on inflammation and liver cell injury, either at one year or with longer treatment duration. This was unexpected, as some studies reported an improvement in inflammation as early as 67 or 12 months of therapy.20, 21 Two randomized, 1-year trials with pioglitazone at 30 mg/day (versus placebo10 or vitamin E8) are available demonstrating significant improvement in steatosis but conflicting results with regard to necroinflammation. Both studies included only nondiabetic patients and some with normal baseline ALT values, contrary to the present trial, where only patients with increased aminotransferases were included and diabetic patients were allowed. Sanyal et al.8 reported no improvement in liver cell injury in a small study with only nine patients with control biopsies per group. In a larger study, comparable in size to the present report, Aithal et al.10 demonstrated a significant improvement in liver cell injury (mostly ballooning) but no improvement in lobular or portal inflammation over the placebo group. Therefore, whether a short (1-year) course of RSG truly improves all aspects of liver injury is not settled, but the present series shows that even longer 2- or 3-year periods of treatment do not provide a clear benefit. Importantly, improvement in insulin sensitivity was well correlated with a reduction in liver fat and in transaminase values; in contrast, there was no such correlation with the course of necroinflammatory lesions. A possible inference is that the correction of steatosis and the accompanying decrease in aminotransferases is directly related to the improvement in insulin sensitivity, whereas necroinflammation and fibrosis might not be reduced significantly by simply correcting insulin resistance. In this trial the absence of fibrosis improvement was documented both on regular pathology and on micromorphometry by assessing the fibrosis stage and the area of fibrosis, respectively. None of these significantly changed during 2 or 3 years of treatment, making it improbable that the lack of effect seen after only 1 year9 could be due to the short treatment period. Of note, several other studies, whether controlled7, 8 or not,20 showed no impact of a 1-year course of either RSG or pioglitazone on liver fibrosis. Only two studies, one of which was uncontrolled,21 showed an improvement of liver fibrosis, both with a very narrow margin of statistical significance.10, 21
Arguments in favor of a specific antifibrotic and antiinflammatory effect of glitazones in the liver have been developed. Such effects, unrelated to the correction of insulin resistance, could justify prolonged therapy. During the in vitro activation of hepatic stellate cells, there is a gradual decline of nuclear peroxisome proliferator activated receptor (PPAR) gamma activity; this can be restored by PPAR gamma agonists with a resultant inhibition of stellate cell activation.22, 23 Moreover, glitazones can prevent liver fibrosis in vivo, in rat fibrosis models unrelated to steatosis accumulation or insulin resistance.24 However, this was not confirmed in all rodent species25 and, importantly, once liver fibrosis is established, glitazone therapy failed to reverse it.26 It is therefore unclear whether glitazones can have a beneficial effect on liver injury independent of their insulin-sensitizing and antisteatogenic effect. The human data presented here are at odds with these experimental data and do not favor such an effect.
The 1-year and the present long-term RSG trials have two potentially important findings. The first is that a large proportion, up to one-half of treated patients,9 are nonresponders, even for easy to attain endpoints such as reduction in steatosis or transaminase values. Second, the correction of insulin resistance might not be sufficient to improve all aspects of liver injury in NASH and therefore to prevent liver disease progression. If confirmed by larger trials this would strongly argue for a tailored therapy that, in addition to insulin sensitizers, would use pharmacological agents with a different mechanism of action. Several studies of hepatoprotective, antiinflammatory or antiapoptotic trials are already undergoing formal testing in clinical trials. One drug might not fit all as far as treatment of patients with NASH.
In conclusion, the limited efficacy of RSG on histological lesions other than steatosis after a 1-year course of therapy and the rapid rebound of serum aminotransferases after cessation of therapy9, 27 argue for prolonged therapy with glitazones in NASH. However, there are side effects and safety issues with these drugs28, 29 and, therefore, the risk/benefit of longer-term treatments should be carefully assessed. Our data provide essential insight on the long-term effects of prolonged RSG therapy by showing that there is no additional histological benefit beyond 1 year of treatment. The present findings but also recent findings of a rapid recurrence of steatosis and inflammation after cessation of glitazone therapy for 1 year or more27 clearly advocate additional studies to devise the best way of using these drugs in patients with NASH.
The members of the LIDO (Liver Injury in Diabetes and Obesity) Study Group are: André Grimaldi, Philippe Giral, Eric Bruckert, Arnaud Basdevant, Karine Clement, Agnès Hartemann-Heurtier, Sophie Gombert, Francine Lamaison, Dominique Simon, Joseph Moussalli, Pascal Lebray, Christiane Coussieu, Djamila Messous, Françoise Imbert-Bismut, Yves Benhamou, Cecilia D'Arrondel, Carole Bernhardt, Isabelle Ravalet, Hôpital Pitié-Salpêtrière; Philippe Podevin, Hôpital Cochin; Christian Boitard, Etienne Larger, Hôpital Hotel-Dieu; Lawrence Serfaty, Chantal Housset, Jacqueline Capeau, Hôpital Saint Antoine all in Paris, France. We thank Janet Ratziu for article editing and Gilles Brami and Denis Vallée from GSK Pharmaceuticals for support during the trial.